The euro introduction and noneuro currencies

Abstract

This article documents the existence of large structural breaks in the unconditional correlations among the US dollar exchange rates of the British pound, Norwegian krone, Swedish krona, Swiss franc and euro during the period 1994 to 2003. Using the framework of Dynamic Conditional Correlation (DCC) models, we find that such breaks occurred both at the time the formal decision to proceed with the euro was made in December 1996 and at the time of the actual introduction of the euro in January 1999. Most correlations were substantially lower during the intervening period. We also find breaks in unconditional volatilities at the same points in time, but these are comparatively of a much smaller magnitude.

Acknowledgements

We thank participants of the conference on ‘Changing Structures in International and Financial Markets and the Effects on Financial Decision Making’ in Venice, Italy, 2–3 June 2005, the 16th (EC)² Conference on ‘The Econometrics of Financial and Insurance Risks’ in Istanbul, Turkey, 16–17 December 2005, seminar participants at Keele University and the University of Leicester, Michael Artis, Claus Brand, Rob Engle, Michael Frömmel, Lex Hoogduin, Franc Klaassen, Denise Osborn, Kevin Sheppard and Genaro Sucarrat for useful comments and suggestions. Any remaining errors are ours alone.

Notes

¹ Topics of particular interest include integration and co-movement of bond and stock markets (Kool, 2000; Morana and Beltratti, 2002; Flavin, 2004; Guiso et al., 2004; Pagano and von Thadden, 2004; Baele, 2005; Kim et al., 2005; Hardouvelis et al., 2006; Bartram et al., 2007), the cost of capital (Hardouvelis et al., 2007), interdependence between US and euro area money markets (Ehrmann and Fratzscher, 2005), convergence of real exchange rates (Lopez and Papell, 2007) and of inflation rates (Honohan and Lane, 2003), trade effects (Micco et al., 2003; Bun and Klaassen, 2007), product market integration (Engel and Rogers, 2004), foreign exchange rate risk exposure (Francis and Hunter, 2004; Bartram and Karolyi, 2006), the behaviour of nominal exchange rates of euro area countries in the run-up to the common currency (Frömmel and Menkhoff, 2001; Bond and Najand, 2002; Wilfling, 2002) and the role of the euro in the foreign exchange market (Detken and Hartmann, 2002; Hau et al., 2002).

² On 1 January 1999 the euro replaced the national currencies of 11 countries: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain. At the time of writing, the euro area comprises 16 countries, now including Cyprus, Greece, Malta, Slovakia and Slovenia.

³ We do not include the Danish krone in the analysis. While being an EU member, Denmark decided not to adopt the euro upon its introduction already in December 1992; a decision that was conformed in a national referendum held on 28 September 2000. Nevertheless, it turns out that the correlation of the Danish krone with the euro has been very close to perfect ever since the euro came into existence. Indeed, since 1 January 1999, the Danish krone has been participating in ERM II with a rather narrow fluctuation and of 2.25%, and effectively has had an almost fixed exchange rate vis-à-vis the euro.

⁴ Although the German mark undoubtedly was the single most important currency in the euro area before 1999, it may not be completely the representative of exchange rate developments in the euro countries. An alternative would be to use the ECU instead of the Deutschmark for the pre-euro period. We do not consider this possibility, however, for the fact that the GBP and SEK were part of the ECU. This obviously may influence results, in particular those pertaining to the correlations of these currencies with the ‘euro’.

⁵ A detailed discussion of this nonparametric volatility and correlation estimator can be found in Hafner et al. (2006).

⁶ Alternative models that allow for time-varying correlations are developed in Pelletier (2006) and Silvennoinen and Teräsvirta (2009), assuming that the correlations switch back-and-forth between a limited number of values, according to an unobserved Markov-switching process or according to the value of observed exogenous variables, respectively. Hafner et al. (2006) generalize the latter approach by combining Equations 1 and 2 with univariate GARCH models for the conditional volatility. We refer to Bauwens et al. (2006a) for a comprehensive survey of multivariate GARCH models. An interesting alternative approach to modelling dependence and changes therein is by means of copulas (see Patton (2006) for an application to exchange rate returns).

⁷ Note that in this case and change at each iteration of the nonlinear optimization procedure such that the unconditional sample volatilities and correlations need to be updated during the estimation process.

⁸ The derivation of SEs of the unconditional covariance matrix that is used for correlation targeting has not been worked out yet, at least not in analytic form. In the univariate context of volatility targeting in a GARCH(1,1) model, Kristensen and Linton (2004) propose to use a Newey–West-type estimator, which we conjecture could also be used in the multivariate context, although its convergence rate is quite slow. We leave improvement of this approach open for future research.

⁹ The analysis was also performed using bivariate models for all possible exchange rate pairs. This led to qualitatively and quantitatively similar results, which are available in full detail upon request.

¹⁰ We should note that in all estimated models we do allow for as many structural changes in μ as there are breaks in the unconditional volatilities and correlations, as discussed below.

¹¹ On the importance of heterogeneity of expectations in foreign exchange markets, see Bollerslev (1990), Lyons (1991, 1996) and Taylor and Allen (1992). For other asset markets, see Easley and O'Hara (1992) and Loretan and English (2000).

¹² Bai and Perron (1998) established the asymptotic properties of this sequential approach, demonstrating consistency and efficiency.

¹³ A detailed analysis of the issue of the appropriate number of breaks in unconditional volatilities and unconditional correlations based on bivariate models is available upon request.

¹⁴ Multivariate GARCH models with smoothly changing unconditional correlations are also considered by Longin and Solnik (1995), Berben and Jansen (2005) and Silvennoinen and Teräsvirta (2009). However, in these studies, this model is developed as an extension of the CCC-model, that is DCC-type dynamics in the conditional correlations are not allowed for.

¹⁵ In the estimation procedure, we enforce that is a genuine correlation matrix by taking the Choleski decompositions of , j = 1, 2, 3, where P_j is a lower triangular matrix and imposing constraints on the nonzero elements of P_j that lead to ones on the diagonal of and automatically give off-diagonal elements between −1 and 1 (see Pelletier (2006) for details).

¹⁶ The accompanying loglikelihood value is equal to −4768, compared to −4793 for the corresponding ‘standard’ DCC model with instantaneous changes in the unconditional volatilities and correlations.

¹⁷ See Cappiello et al. (2006) for other generalizations of the DCC model.

¹⁸ These may be compared with the square root of the estimate of γ in the DCC model with two volatility and correlation breaks, which is equal to 0.152.